Like the article said, they're counting by integrated graphics generations now, so whether or not the CPU architecture remains the same between different GPU generations is irrelevant now.
The article is wrong about that.
If they did that, they couldn't count the first generations of Core i7, because they didn't have a GPU at all.
Case in point:
http://ark.intel.com/products/52214/Intel-Core-i7-2600K-Processor-8M-Cache-up-to-3_80-GHz
This is Sandy Bridge. Clearly '2nd generation' as you can read on Intel's own page.
But Sandy Bridge was the first Core i7 to have a GPU onboard. So how can it be second generation if they count by GPU?
It just so happens that Intel updates the GPU at every new generation as well, now that it is integrated.
Their example i5 655K is not marked as 'first generation':
http://ark.intel.com/products/48750/Intel-Core-i5-655K-Processor-4M-Cache-3_20-GHz?q=core i5 655K
It is part of 'Previous generation Core i5':
http://ark.intel.com/products/family/59141/Previous-Generation-Intel-Core-i5-Processor#@Desktop
Then again, looking at 'Previous generation Core i7' also lists all the Nehalems:
http://ark.intel.com/products/family/59143/Previous-Generation-Intel-Core-i7-Processor#@Desktop
Some of these have GPUs, most don't.
Besides, how does it even make sense to count CPU generations by the GPU? There is absolutely no reason why Intel would do that.
The problem is that the article tries to capture multiple families into a single number as I already said.
That's the same error as benchmarks trying to capture the performance of a CPU in a single figure. It doesn't work that way.
The article is wrong about many things anyway...
For example:
Intel didn’t have the capability to actually put the graphics into the CPU die itself back then.
Erm, sure it did. It made CPUs that were much larger than the CPU and GPU of the i3/i5 combined on a single die, such as the 6-core Westmere-EX (which does not have a GPU... yet is the same generation).
The capability was there. Economically it would not have been the best choice however.
I guess this guy probably also thinks that all Pentium Ds were two-die. Most people forget that the original Pentium D was the 800-series, and it was indeed a single-die solution ('native dualcore', to speak with AMD's hollow marketing).
So yes, Intel could surely build them on a single die. But yield-wise it made more sense to use separate dies at the time, so they moved to that.
